Re: THEORY: Dirk on ambisyllabicity

On Sat, 14 Oct 2000, And Rosta wrote:

> Dirk:
> > On Sun, 8 Oct 2000, And Rosta wrote:
> >
> > > FWIW (= not very much), my conclusion is that timing units are separate
> > > from syllable structural units. E.g. (using O = onset, R = rhyme, T =
> > > timing units):
> > >
> > >      O     R   O R          O     R     O R
> > >     / \   / \ /  |         / \   / \    | |
> > >    T   T T   T   T        T   T T   T   T T
> > >    |   | |   |   |        |   | |    \ /  |
> > >    s   t i   l   i        s   t i     l   i
> > >
> > >        [stIli]              [stIlli]
> >
> > I would agree. I would go further and say that onsets are not involved
> > in timing relations, and that such relations are best expressed by
> > morae.
> >
> >             s     s              s     s
> >            /|    /|             /|\   /|
> >           / m   / m            / m m / m
> >          /  |  /  |           /  | |/  |
> >         st  I l   i          st  I l   i
>
> Don't you mean this? :-
>
>              s     s              s     s
>             /|    /|             /|\   /|
>            / m   / m            / m m / m
>           /  |\ /  |           /  | |/  |
>          st  I l   i          st  I l   i
Well, no. To be consistent with my anti-ambisyllabcity story I suppose
I *really* meant

               s   s                s     s
              /|\  |               /|\   /|
             / m m m              / m m / m
            /  | | |             /  | |/  |
           st  I l i            st  I l   i

with the distinction between geminate and non-geminate apparent only
in the association properties of [l]; in the first, [l] is associated
only to the first syllable, while in the second it is associated to
both. This would then not agree so much with your parse given above.
However, if ambisyllabicity were seriously considered, then your
revision of my [stIli] seems sensible.

> Apart from this, I agree with everything you say. I considered using
> diagrams like yours, but hesitated for several reasons.
>
> (1) I'm not aware of positive arguments for a s node (though in effect
> O and s seem to be notational equivalents).
Did you learn phonology from John Harris or Jonathan Kaye? Your
representations remind me of Government Phonology. I regret to say
that I don't understand the theory well enough to offer a critique,
but one thing springs to mind. If there is no s (=syllable) node, then
how does the theory rule out consecutive Os? Consecutive Rs are not a
problem, since under a more traditional account of prosodic structure,
syllables need not have onsets; but they must have at least a nucleus,
and hence a rhyme. So the presence of a syllable node is useful at
least to account for the absence of consecutive onsets without
rhymes. However, under a theory which does away with the syllable,
there doesn't seem to be anything to rule out consecutive Os other
than bald stipulation.

As for 'O' and 's' being notational equivalents, I'm afraid don't see
it. Doubtless this has to do with particulars of Government Phonology
that I'm not familiar with. However, since our discussion seems to be
about the multiple linking of a consonant between a rhyme and a
following onset, I don't know that it is crucial to insist on an s
node.

> (2) For most dialects of English there is a rule that the nucleus and
> the coda can each branch (into two segments each) only if the coda is
> coronal. (For A-lengthening dialects, the rule holds only for nasal
> + plosive codas, because long A introduces exceptions elsewhere.) It's
> not clear to me how this contraint would be formulated in a mora-based
> analysis if it eschews nucleus and rhyme nodes. [Incidentally, if
> /ink/ and /eip/ are both bimoraic, to which mora does the medial segment
> in each word belong?]
What seems to be at issue here is the fact that the rhyme may contain
up to four timing positions. Take the word 'pint'; I'm guessing that
GP would parse this as follows:

        O    R
        |   / \
        |  N   C
        | / \ / \
        x x x x x
        | | | | |
        p a I n t

The problem for moraic phonology is two-fold: 1) how many morae
are present? 2) how are morae assigned? The first is a problem, since
there seems to be a well-founded principle limiting syllables to two
morae, but there are twice that many segments in the rhyme of 'pint'.
The second is a problem since there does not seem to be a principled
way to parcel out the segments to the morae. My guess is that the
first principle must be relaxed to allow at least a trimoraic
syllable. Under this proposal, final coronals are never moraic when
they follow another coronal. This sounds stipulatory, but I believe
it's the route taken by Hammond (1999). Accordingly, 'pint' would
receive the following moraic parse:

           s              s
         /| |\         / /|\ \
        / m m m   or  / m m m \
        | | | /\      | | | | |
        p a I n t     p a I n t

With the final [t] sharing the third mora, or alternatively, linked
directly to the s-node (I'm not aware of arguments which would compell
one structure over another).

The examples you cite, [ink] and [eip], would also be trimoraic.

> (3) I'm not sure whether the the moraic analysis allows for (A)
> segmentally empty positions such as would account for liaison blocking
> in French (H-aspire, and ouatte/watt [ex. borrowed from someone else
> (J-R Vergnaud?)], and perhaps for AN-cliticization in English and (B)
> for unattached segments, such as word-initial geminates in 'classical'
> Calabrese (which are realized as geminates only when following a
> vowel-final word).
Moraic phonology allows for both types of unaffiliation. The analysis
of compensatory lengthening proposed by Bruce Hayes in his 1989 LI
article makes use of (temporarily) unaffiliated morae. The following
two rules are used to account for the historical development of Latin
[ka:nus] 'gray' from earlier *[kasnus]. First a rule of /s/ deletion:

        s -> 0 / _ [+son,+ant]

This rule only applies on the segmental tier and doesn't affect
prosodic structure. Compensatory lengthening then involves the
spreading of the vocalic features of the immediately preceding vowel
to the now empty mora:

        m m    m m
        | | -> |/
        V 0    V

So the derivation goes like this:

          s     s
         /|\   /|\
        / m m / m m
        | | | | | |
        k a s n u s

/s/ Deletion:

          s     s
         /|\   /|\
        / m m / m m
        | | | | | |
        k a 0 n u s

Compensatory Lengthening:

          s     s
         /|\   /|\
        / m m / m m
        | |/  | | |
        k a   n u s = [ka:nus]

I also use empty slots (not moraic, though) to account for the
Geminating series in Shoshoni phonology. Briefly, voiceless stops are
geminated following certain morphs:

        kuna  'firewood'
        kunappai 'have firewood'

        cf

        kynu  'grandfather (FaFa)'
        kynuBai 'have a grandfather'

In the first stem, there is an empty root node specified [+cons]. It
receives its featural content from a following consonant. This creates
a geminate, which is immune from lenition. When a consonant is not
available, the empty root node remains unrealized, having no segmental
material to draw from. In the second example, the stem is vowel-final;
the initial consonant of the suffix is thus properly intervocalic and
lenites.

There are also examples of segments which are not linked into prosodic
structure; Slavic jers have received this kind of treatment.
Basically, these "latent segments" are pronounced by creating prosodic
positions under certain conditions to allow the realization of their
featural material. If the specific conditions are not met, the latent
segments remain unpronounced.

Dirk

--
Dirk Elzinga
dirk.elzinga@m.cc.utah.edu